Professional Documents
Culture Documents
A Concrete Approach to
Classical Analysis
Solutions to Exercises
February 29, 2012
Springer
Contents
5 Differential Calculus on R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
6 Integral Calculus on R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
7 Differential Calculus on Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
1.1. (i) Suppose that the set in the left-side contains an element, say x. Then
x belongs to A, not to B, and to B. We get a contradiction, hence the set
in the left-side contains no element.
(ii) Obviously, A ∩ B ⊂ A \ (A \ B). If there exists an x ∈ A \ (A \ B), then
x ∈ A and x ∈ / A \ B. It follows that x ∈ B, so that x ∈ A ∩ B.
(iii) Both sides of this identity contain elements belonging either to A or to
B.
(iv) Both sides of the identity contain elements in A and neither in B nor in
C.
(v) The three members of this double identity contain elements belonging to
A and C, but not to B.
x ∈ (A ∪ B) \ C ⇐⇒ (x ∈ A or x ∈ B) and x ∈/C
⇐⇒ (x ∈ A and x ∈/ C) or (x ∈ B and x ∈
/ C)
⇐⇒ x ∈ (A \ C) ∪ (B \ C).
(ii) We have
x ∈ (A ∩ B) \ C ⇐⇒ (x ∈ A and x ∈ B) and x ∈
/C
⇐⇒ (x ∈ A and x ∈ / C) and (x ∈ B and x ∈
/ C)
⇐⇒ x ∈ (A \ C) ∩ (B \ C).
6 1 Sets and Numbers
(A ∪ C) \ B = (A \ B) ∪ (C \ B) ⊂ (A \ B) ∪ C.
1.5. (i)
x ∈ ∪∞
n=0 Bn ⇐⇒ ∃n ∈ N, x ∈ Bn ⇐⇒ ∃n ∈ N, i ∈ {0, . . . , n}, x ∈ Ai
⇐⇒ x ∈ ∪∞
n=0 An .
(ii)
x ∈ ∩∞
n=0 Bn ⇐⇒ ∀n ∈ N, x ∈ Bn ⇐⇒ ∀n ∈ N, i ∈ {0, . . . , n}, x ∈ Ai
⇐⇒ x ∈ ∩∞
n=0 An .
1.6. (i)
x ∈ ∪∞ ∞ ∞
m=0 (∩n=0 Am,n ) =⇒ ∃m0 ∈ N, x ∈ ∩n=0 Am0 ,n
=⇒ ∃m0 ∈ N, ∀n ∈ N, x ∈ Am0 ,n
=⇒ x ∈ ∩∞ ∞ ∞ ∞
n=0 (∪m=m0 Am,n ) ⊂ ∩n=0 (∪m=0 Am,n ).
(iv) x ∈ {B =⇒ x ∈
/ B =⇒ x ∈
/ A =⇒ x ∈ {A.
1.9. (i)
1.12. (a) (i) =⇒ (ii) For every ξ ∈ C we have that (f ◦ g)(ξ) = (f ◦ h)(ξ),
i.e., f (g(ξ)) = f (h(ξ)). Because of f is one-to-one, it follows that g(ξ) = h(ξ)
and since ξ was taken arbitrarily in C, we have that g = h.
(ii) =⇒ (i) Suppose that f is not one-to-one. Then there exist x, y ∈ A, with
x ̸= y and f (x) = f (y). Consider C = {u, v} and the functions g, h : C → A
8 1 Sets and Numbers
S = {1} =⇒ A = {S},
S = {1, 2} =⇒ A = {S, {{1}, {2}}},
S = {1, 2, 3} =⇒ A = {S, {{1}, {2}, {3}}, {{1, 2}, {3}}, {{1, 3}, {2}},
{{2, 3}, {1}} },
S = {1, 2, 3, 4} =⇒ A = {S, {{1}, {2}, {3}, {4}}, {{1, 2, 3}, {4}},
{{1, 2, 4}, {3}}, {{1, 3, 4}, {2}}, {{2, 3, 4}, {1}}, {{1, 2}, {3, 4}},
{{1, 2}, {3}, {4}}, {{1, 3}, {2, 4}}, {{1, 3}, {2}, {4}}, {{1, 4}, {2, 3}},
{{1, 4}, {2}, {3}}, {{2, 3}, {1}, {4}}, {{2, 4}, {1, 3}}, {{3, 4}, {1}, {2}}}.
Thus the equalities in (1.11) of the statement hold. Some results can be ob-
tained by Mathematica r , Figure 1.1.
One can also try using the next Mathematica r command.
Table[BellB[n], {n, 0, 4}]
Now suppose S = {a1 , a2 , . . . , an , an+1 }. Pick up( an
) arbitrary element in S,
say an+1 . Select k elements in S \ {an+1 } (in nk ways) and denote by A
the set of remaining n − k elements. By A one has Bn−k partitions. Thus we
get (1.12).
1.17. Every x ∈ A belongs to at least a set Ai . The set of indices of the sets
Ai containing x is a nonempty subset to {1, 2, . . . , n}. So, there are 2k − 1
possibilities for x belonging to at least a set Ai . Since each x ∈ A belongs to
at least a set Ai , the conclusion follows.
10 1 Sets and Numbers
H* Set partitions*L
SetPartitions@4D
8881, 2, 3, 4<<, 881<, 82, 3, 4<<, 881, 2<, 83, 4<<, 881, 3, 4<, 82<<,
881, 2, 3<, 84<<, 881, 4<, 82, 3<<, 881, 2, 4<, 83<<, 881, 3<, 82, 4<<,
881<, 82<, 83, 4<<, 881<, 82, 3<, 84<<, 881<, 82, 4<, 83<<, 881, 2<, 83<, 84<<,
881, 3<, 82<, 84<<, 881, 4<, 82<, 83<<, 881<, 82<, 83<, 84<<<
H* Bell numbers*L
BellB@4D
15
1.18. By induction.
Applying the above relation to each nonzero term in the left-side of (1.13),
we get the result.
Since the coefficient of the leading term is 1 and all the other coefficients are
integers, the rational roots are integers. If t is odd, we get a contradiction.
If t is even, the leading term is divisible by 4, while the product is divisible
only by 2. We get a contradiction again. Thus the polynomial equation has
no integer root. So it has no rational root. ⊓ ⊔
1.22. Suppose that |A| = n. Then P(A) is equivalent with the set of n-term
sequences of 0 and 1.
1.1 Solutions 11
√
1.23. Multiply (1.14) by x and add −1. Then (1.14) is equivalent to
√ √ √ √
−( x + 1 − x)2 < 0 < ( x − x − 1)2 ,
1.24. Since the inequality is symmetric in the variables, we may assume that
x ≥ y ≥ z. Then we can write
1.25. By induction.
1.26. For x = −1 the left-hand side of (1.16) in the statement is null while
the right-hand is non-positive. So we may suppose that x > −1. But for x >
−1 inequality (1.16) follows immediately from inequality (1.15) considering
xi = x, i = 1, 2, . . . , n.
1.27. (i) In (1.16) we set x = 1/n to get the left hand side inequality. To get
the other one we apply the Newton formula. If an = (1 + 1/n)n , then
( ) ( ) ( )
n 1 n 1 n 1
an = 1 + + + ··· +
1 n 2 n2 n nn
n(n − 1) 1 n(n − 1)(n − 2) . . . 2 · 1 1
=1+1+ 2
+ ··· + n
( 2! ) n ( ) (n! )( n )
1 1 1 2 n−1 1
=1+1+ 1− + ··· + 1 − 1− 1−
n 2! n n n n!
1 1 1
< 1 + 1 + + + ··· +
2! 3! n!
Since n! ≥ 2n−1 , for every n ≥ 2, it follows that
( )
1 1 1
2 < an < 1 + 1 + + 2 + · · · + n−1 < 3.
2 2 2
or
√ √ √( )n
n+1
n+1 n(n+1) (n + 1)n n(n+1) 1 1
√
nn
= = 1+ .
nn+1 n n
12 1 Sets and Numbers
2√
k x1 + x2 + · · · + x2k
x1 x2 . . . x2k ≤ . (1.1)
2k
Then
√
√
2k+1 2√
k √
k
x1 x2 . . . x2k+1 = x1 . . . x2k 2 x2k +1 . . . x2k+1
2√
k √
k
x1 . . . x2k + 2 x2k +1 . . . x2k+1
≤ .
2
Using (1.1), we get that (1.17) holds for any m equal to a power of 2.
It remained the case of unrestricted m. For, suppose that m is not a power
of 2. Take a k ∈ N∗ such that m < 2k and denote
x1 + x2 + · · · + xm
l= (> 0).
m
Set
xm+1 = · · · = x2k = l
and consider inequality (1.1). Then we write
2√ ml + (2k − m)l
x1 x2 . . . xm · l(2 −m)/2 ≤
k k k
=l
2k
2√
k k
x1 x2 . . . xm ≤ lm/2 ⇐⇒ x1 . . . xm ≤ lm .
x1 , . . . , xm > 0, x1 · · · xm = 1 =⇒ x1 + x2 + · · · + xm ≥ m. (1.2)
x1 x2 · · · xn xn+1 = 1.
x1 , x2 , . . . , xn−1 , xn · xn+1 .
Based on the induction hypothesis, (1.3), we get the following lower estimate
x1 + x2 + · · · + xn−1 + xn xn+1 ≥ n.
Further
Thus we infer that implication (1.3) holds for every m ∈ N∗ . Hence the mean
inequality is proved on this the second way as well.
∑
n
1 ∑n ∑ 1
n
(n − 1)s = (s − ak ) ≥ n2 .
s − ak s − ak
k=1 k=1 k=1
∑ (√ √ )2
ti /tj − tj /ti < 1.
1≤i<j≤n
1.35. (i) First approach. We just remark that (1.21) follows from Lagrange
identity (1.20).
Second approach. Consider the following inequalities
1.36.
∑
n ∑
n ∑
n ∑
n
a2k ≤ an ak and b2k ≤ bn bk .
k=1 k=1 k=1 k=1
1.37. We prove the second inequality. The first inequality can be proved
similarly. ∑n ∑n
The sum ( k=1 ak ) ( k=1 bk ) contains n2 terms of two kinds
∑n
(i) n terms of the form ak bk . Their sum is written as k=1 ak bk ;
n(n − 1) terms of the form ai bj with i ̸= j. Their sum is written as
(ii) ∑
i̸=j ai bj .
The terms in (ii) can be grouped as ai bj + aj bi . For every such pair we can
write
1.39. (i) The inequality can be proven by induction. For n = 1 one has an
identity. For n = 2 one has
∏
2 ∑
2
(1 − ak ) = 1 − a1 − a2 + a1 a2 ≥ 1 − a1 − a2 = 1 − ak .
k=1 k=1
∏
m+1 ∏
m ∏
m ∏
m
(1 − ak ) = (1 − am+1 ) (1 − ak ) = (1 − ak ) − am+1 (1 − ak )
k=1 k=1 k=1 k=1
∏
m ∑
m+1
≥ (1 − ak ) − am+1 ≥ 1 − ak .
k=1 k=1
1.41. If any ak or bk is zero, the inequality is trivial. Suppose that all entries
are positive. By the mean inequality, we have
a1 a2 an
√ + + ··· +
a 1 a 2 a n a + b1 a2 + b2 an + bn
n ... ≤ 1
a1 + b1 a2 + b2 an + bn n
√ b1 b2 bn
+ + ··· +
b1 b2 b n a + b a + b a n + bn
n
... ≤ 1 1 2 2
.
a1 + b1 a2 + b2 an + bn n
Add side by side the two inequalities and the result follows.
1.42. First approach. The identities follow by the Moivre formula from
(√ ( π π ))n ( nπ nπ )
(1 + i)n = 2 cos + i sin = 2n/2 cos + i sin .
4 4 4 4
Second approach. One proves it by induction.
One ca try using the following Mathematica commands r
TraditionalForm[Sum[(-1)^k Binomial[n, 2*k], {k, 0, n/2},
Assumptions \[RightArrow] n \[Element] Integers &&
n \[GreaterSlantEqual] 0]]
TraditionalForm[Sum[(-1)^k Binomial[n, 2*k + 1], {k, 0, n/2},
Assumptions \[RightArrow] n \[Element] Integers &&
18 1 Sets and Numbers
n \[GreaterSlantEqual] 0]
Hence
⌊ ⌋ (⌊ ⌋ ) ⌊ ⌋ (⌊ ⌋ )
n+2 n+2 n+1 n+1
|Sn | = (n + 1)2 + −1 + −1
2 2 2 2
⌊ n ⌋ (⌊ n ⌋ ) ⌊ n − 1 ⌋ (⌊ n − 1 ⌋ )
= (n + 1)2 + +1 + +1 .
2 2 2 2
1.44. We use the theorem that every nonempty subset of N has a smallest
element in respect to the natural ordering, i.e., N is well-ordered.
Suppose there exists a ∈ N such that f (a) ̸= g(a). Then the set
A = {n | n ∈ N, f (n) ̸= g(n)}
From here it follows that b ∈ A and g(b) < g(a) = f (b) < f (a). Now
the minimality of g(a) is violated. So our assumption does not hold and we
conclude that f = g.
1.45. We are looking for functions satisfying the condition that there exists
a nonnegative real m so that |f (k)| ≤ m for all k ∈ Z.
Suppose there exists f a solution. For n = k = 0, we have
From here we immediately have that f (n) = 1, for all n ∈ Z and this f is a
solution of the exercise.
Suppose f (1) = −1. Then for n = 1, we have
Immediately we have that f (k) = (−1)k , for all k ≥ 0, and so f (k) = (−1)k ,
for all k ∈ Z. One can check that this function is a solution of the exercise.
Suppose f (1) = 0. For n = 1, we have
1.46. We start by proving that f (0) < f (1) < f (2) < . . . by induction. Let
Pn be the statement that f (n) is the smallest element of {f (n), f (n+1), . . . }.
For m > 0, by assumption, f (m) > f (s), where s = f (m − 1). So f (m) is
not the smallest element in {f (0), f (1), f (2), . . . }. This set being a nonempty
subset of N, has a smallest element. Hence the unique smallest element is
f (0). Thus P0 is true.
Suppose Pn is true. Take m > n + 1. Then m − 1 > n and so f (m − 1) >
f (n). We also have that
From (1.4) and (1.5) we conclude that if f satisfies the assumptions, then
We can check easily that this function satisfies all assumptions. So, the
only one solution to our exercise is f (n) = n, for all n ∈ N.
1.47. Under the form in the statement, the exercise has no solution since
in the right hand side occurs 00 , which is rejected. Instead some cases are
1.1 Solutions 21
interesting, the method that we use here in the main case is based on the
separation of variables.
(a) Consider the simplest case
where α ̸= 0.
For y = 0, we have
Thus either f (0) = 0 or f (x) = 1, for all x ∈ [0, ∞ [ . We reject the case
f (x) = 1, for all x ∈ [0, ∞[ , it does not satisfy (1.7). For x = 0, we have
The left hand side (1.24) is symmetric in x and y, so the right hand side has
to be as well. Therefore
f (x/2) f (y/2)
= α .
x −x
α β y − yβ
22 1 Sets and Numbers
So the function
f (x/2)
x 7−→
xα − xβ
is constant on ]0, 1[ ∪ ]1, ∞ [ . This implies that there is a real constant c such
that
Under these conditions the functional equation (1.24) has the form
It follows that c = 0. So
1.49.
Consider A = ]0, 1]. Show that int A = ] 0, 1[ , cl A = [0, 1], and A′ =
[0, 1].
Bbc
C
bc
x
x
bc bc
A
O
2.1. R is a vector space over Q because each rational number is a real number
and R is a commutative field. Q is not a vector space over R because the
product between a scalar (in Q ) and a vector (in R) generally does not belong
to Q.
2.2. The sum of two polynomials with real coefficients is a polynomial with
real coefficients. The product of a real number by a polynomial with real coef-
ficients is a polynomial with real coefficients. The null element is the constant
null polynomial.
2.3. The first answer follows from the previous exercise. Since {1, x, . . . , xn }
is a basis, the dimension of this vector space is n + 1.
2.4. No. For example, the sum of the first degree polynomials p(x) = x by
q(x) = −x does not belong to the set of first degree polynomials.
Hence ax + by ∈ c0 .
2.7. Suppose that the claim is not true. Then for every positive integer k
there exists a system {x1k , x2k , . . . , xnk } of linear dependent vectors ful-
filling ∥xi − xik ∥ < 1/k, i = 1, 2, . . . , n. Then there exist some coefficients
c1k , c2k , . . . , cnk so that
2.9. int A = ]0, 1[ , cl A = [0, 1] ∪ {2}, A′ = [0, 1], and fr A = {0, 1, 2}.
3.1. Consider the sequence (an ) of all natural numbers, but not perfect
squares, i.e., a1 = 2, 2 = 3, a3 = 5, a4 = 6, . . . . Set
m
xk,m = (ak )2 , for k = 1, 2, . . . , and m = 0, 1, 2, . . . .
Then
xk,m+1 = x2k,m
3.2. Indeed, it is enough to remark that the sequence has positive terms, so
it is monotonically increasing, but divergent. To check this the last property
it is enough to see that it is not a Cauchy one. The sequence (an ) is not
fundamental because
3.3. Suppose (an ) is bounded. Then there exists a positive M such that
|an | ≤ M, for every n. From the hypothesis we infer that the open intervals
]an − 1/(2n), an + 1/(2n)[ are disjoint and their union satisfies
28 3 Sequences and Series
3.7. We remark that (xn ) has positive terms and is increasing since for all
n ≥ 1,
√
√ √
x2 = a + a > a = x1 , (xn+2 − xn+1 )(xn+2 + xn+1 ) = xn+1 − xn .
and we note that for every n ∈ N∗ yn = xn /e, i. e., (xn ) and (yn ) converge
or diverge simultaneously. We also remark that (yn ) is increasing.
From the hypothesis it follows that there exists an n0 ∈ N∗ such that for
n ≥ n0 ,
n
(1/n) ln(ln an ) < ln 2 ⇐⇒ an < e2 ⇐⇒ bn < 1.
Consider a = max{b1 , b2 , . . . , bn0 , 1} and define the following sequence
√
√ √ √
z1 = a, z2 = a + a , . . . , zn+1 = a + zn , . . . .
0 ≤ bn < bn+1 < an+1 < an , 0 < an+1 − bn+1 < (an − bn )/2, n ≥ 1,
3.11. All the terms are positive. Suppose there exists n ∈ N∗ such that
an > bn . Then
an+1 2bn
= < 1 =⇒ an+1 < an ;
an an + bn
( )2
b2n+1 2an bn bn+1 2an
= =⇒ = > 1 =⇒ bn+1 > bn ;
bn an + bn bn an + bn
√
bn+1 an+1
= > 1 =⇒ an+1 > bn ;
bn bn
√
an+1 an+1 an+1
=√ = > 1 =⇒ an+1 > bn+1 .
bn+1 an+1 bn bn
From the above inequalities we conclude that (an ) decreases, (bn ) increases,
and an < bn , for all n ∈ N∗ . Hence the two sequences converge. Passing to
the limit in the recursion of bn , we get that the two limits coincide. Why the
common limit is equal to 2π is shown at page 356.
1 + 2 2 + · · · + nn 1 + n + n2 + · · · + nn nn+1 − 1 1 n→∞
1≤ ≤ = −−−−→ 1.
n n n n n − 1 nn
Thus we get that the limit is 1.
For this limit we can also use Theorem 1.22 of Stolz–Cesaro.
3.13. (a) We note that xn > 0 for all n ∈ N∗ \ {1} and xn < 1 for all n ∈ N∗ .
We can improve the upper of the sequence since
√ √
−1 + 5 −1 + 5
xn < =⇒ xn+1 < , for all n ∈ N∗ .
2 2
Using the previous result, it is clear that xn+1 > xn . Thus the sequence
is monotone and bounded, therefore√convergent. Passing to the limit in the
recurrence, we find the limit (−1 + 5)/2.
(b) It converges to 5. √
(c) It converges to (−1 + √5)/2.
(d) It converges to (−1 + 5)/2.
(e) If p = 0, it results the constant sequence xn = 1, for all. So the sequence
converges to 1. If p = 1, the general term of the sequence is equal to xn+1 =
n+a, thus the sequence diverges. If p ∈ ]0, 1 [ , the general term of the sequence
is equal to xn+1 = 1 + p + · · · + pn−1 + apn , thus the sequence converges to
1/(1 − p).
3.14. We note that the two sequences are unbounded. Then there exist n1 , n2
satisfying |an1 − an2 | > 2 (since otherwise (an ) is bounded). Further, there
exists n3 such that |bn1 − bn3 | > 1 and |bn2 − bn3 | > 1 (since otherwise the
sequence (bn ) is bounded). Now, if |an1 −an3 | > 1, then n = n1 and m = n3 .
If |an1 − an3 | ≤ 1, then |an2 − an3 | > |an1 − an2 | − |an1 − an3 | > 1, and hence
n = n2 and m = n3 .
3.15. Note that always lim inf n→∞ an ≤ lim inf n→∞ ak n for any subse-
quence (ak n ). We have lim inf n→∞ (an + bn ) = limn→∞ (ap n + bp n ) and
lim inf n→∞ ap n = limn→∞ am p n . By the previous note
Now we write
we have
lim inf (an + bn ) − lim sup bn = lim inf (an + bn ) + lim inf (−bn )
n→∞ n→∞ n→∞ n→∞
Thus the second inequality in (3.83) holds. Similarly can be proved (3.84).
Denote
ηn = (|b1 | + |b2 | + · · · + |bn |)/n − |b|
and choose a positive ε. Then there is a natural N so that for n > N,
|an | < ε/(4b). Then for n > N,
|ηn | < |b| and M (|a1 | + · · · + |aN |)/n < ε/2, for all n > ν.
where
dn = (α1 bn + α2 bn−1 + · · · + αn b1 )/n.
Since αn → 0, we have that dn → 0, too. We apply Corollary 1.5 to (3.1)
and we find that cn → ab, as n → ∞.
3.17. We denote by M the set of constants α for which the above limit exists
and it is finite. Remark that M ̸= ∅ since by Proposition 1.18, 1 ∈ M. Now
γ, α = 1,
1 1 n→∞
1 + + · · · + − ln n + (1 − α) ln n −−−−→ +∞, α < 1,
2 n
−∞, α > 1.
Hence M = {1}.
Suppose not all the members are equal. Suppose k ∈ {2, . . . , n − 1} and
a1 < ak , ak > an . Then
and thus
a2k = ak−1 ak+1 .
One has ak ≥ ak−1 and ak ≥ ak+1 . Thus we get ak−1 = ak = ak+1 . Using
the same argumentation, we find that all the elements of the set are equal,
which is a contradiction. Thus the maximum is equal to a1 or an . Both results
are possible. Suppose a1 = 1, a2 = 2, a3 = 3. Then max{a1 , a2 , a3 } = a3 .
Suppose now a1 = 3, a2 = 2, a3 = 1. Then max{a1 , a2 , a3 } = a1 .
3.19. From xn+1 = xn (xn −1)+1 we have that xn > 1, for all n ≥ 1 and from
xn+1 −xn = (xn −1)2 > 0 we conclude that the sequence is increasing. Suppose
it is bounded. Then it converges and let limn→∞ xn = α > 1. From the
recurrence we find out that α = 1, a contradiction. Thus (xn ) monotonically
diverges to ∞. Since xn+1 − 1 = xn (xn − 1), then
1 1 1 1 1 1 1
= = − =⇒ = − .
xn+1 − 1 xn (xn − 1) xn − 1 xn xn xn − 1 xn+1 − 1
Thus the general term of the sequence of partial sums can be written as
1 1 1
sn = + + ··· +
x1 x2 xn
1 1 1 1 1 1
= − + − + ··· + −
x1 − 1 x2 − 1 x2 − 1 x3 − 1 xn − 1 xn+1 − 1
1 1
= − .
x1 − 1 xn+1 − 1
n→∞
Since (xn ) diverges to ∞, we conclude sn −−−−→ 1/(a − 1) and therefore
∞
∑
1/xn = 1/a − 1.
n=1
3.20. The sequence (xn ) is convergent. Let x = lim xn . The general term
of the series is nonnegative. Thus the series is convergent if and only if the
sequence of partial sums is bounded. The general terms of the sequence of
partial sums can be written as
3.21. (i) It diverges because is the generalized harmonic series with p = 1/2.
(ii) It converges. One can check it by the root of ratio tests.
(iii) It diverges.
(iv) It diverges because is the generalized harmonic series with p = 1/3.
(v) It converges based on the comparison, ratio, or root test. Also try the
Mathematica r command
Sum[(n + 1)/( n^2 3^n), {n, 1, Infinity}]
(vi) It converges to 5 sin(x)/(26−10 cos x). Try the Mathematica r command
Sum[Sin[n * x]/5^n, {n, 1, Infinity}]
(vii) It converges to 1. Try the Mathematica r command
Sum[2/3^n, {n, 1, Infinity}]
3.22. (a) Its domain of convergence is the set ]1, ∞[ , according to Theorem
3.10. On this set the series converges even absolutely. Try the Mathematica r
commands
3.1 Solutions 35
SumConvergence[1/n^x, n]
So, for |x| < 1, the series converges absolutely. For x = −1, the series con-
verges by Leibniz Theorem. For x = 1, the series is the harmonic series, thus it
diverges. For |x| > 1, the necessary condition is not satisfied, thus it diverges.
(e) We have
√
n |x − 3|
lim sup |(x − 3)n /nn | = lim sup = 0.
n→∞ n→∞ n
Thus the given series converges for all x ∈ R. Try also the Mathematica r
command
SumConvergence[(x - 3)^n/n^n, n]
(f) By the ratio test we have
|x|, |x| < 1, x ̸= 0
xn+1 (1 + x2n ) 1 + x2n
lim sup = |x| · lim = 1, |x| = 1
(1 + x2n+2 )xn n→∞ 1 + x2n+2
1
n→∞
, |x| > 1.
|x|
(g) For x = 0, the series diverges. The ration test gives no information re-
garding the convergence. By the Raabe–Duhamel test, i.e., Theorem 3.14, we
have
( )
n! (x2 + 1)(x2 + 2) · · · (x2 + n + 1)
lim inf n · − 1
n→∞ (x2 + 1)(x2 + 2) · · · (x2 + n) (n + 1)!
= x2 .
Thus the series converges for x2 > 1, that is x < ∑ −1 or x > 1. It diverges
for −1 < x < 1. If x2 = 1, the series looks like 1/(n + 1), so diverges.
Hence the series converges if and only if x2 > 1. Try also the Mathematica r
commands
$Assumptions = {x \[Element] Reals};
SumConvergence[n!/Product[x^2 + i, {i, 1, n}], n]
3.25. The sequence (an (x)), where an (x) = 1/(x2 + n), has positive terms
and converges monotonically to 0. Since |an (x) − 0| ≤ 1/n → 0 as n → ∞,
we conclude that the sequence (an (x)) converges uniformly to 0 on R. By
Theorem
∑∞ 4.3 we have that the series converges uniformly on R. The series
2
n=1 1/(x + n) diverges for each real x, since comparing with the harmonic
series, the comparison test supplies the answer.
4
Limits and Continuity
4.1. Consider p > 0 since otherwise the necessary condition for convergence
is not satisfied. Denote f (x) = x−p , x > 0. We have that ex · f (ex )/f (x) =
xp /e(p−1)x . Thus for p > 1, limx→∞ ex · f (ex )/f (x) = 0 < 1, and we conclude
the series converges. For 0 < p < 1, ex · f (ex )/f (x) ≥ 1, and we conclude the
series diverges. The case p = 1 has been already discussed.
Further |ρ(cos3 θ + sin3 θ)| ≤ 2ρ. Hence the function is continuous on the real
plane.
4.3. Note the null function and the identity function are solutions to (4.54).
Thus the set of solutions is nonempty. Note also that a solution necessarily is
an odd function. We are looking for the whole set of solutions.
Taking x = y = 0, we find φ(0) = 2φ(0), i.e., φ(0) = 0. By induction
one can show that φ(n) = nφ(1), n ∈ N∗ . Then from φ(1) = φ (m/m) =
mφ (1/m) it follows φ (1/m) = 1/mφ(1) for m ∈ N∗ .
By induction one can further show that φ(p/q) = (p/q)φ(1), p ∈ Z,
q ∈ Z \ {0}. Now we know the representation of the solution if we restrict the
exercise to the system of rational numbers.
Consider an arbitrary but fixed x ∈ R \ Q. Then there exists a sequence
(qn ) of rational numbers such that qn → x. By the continuity hypothesis one
can conclude that
40 4 Limits and Continuity
From the last equality taking into account the continuity assumption we infer
x
that φ(x) = (φ(1)) , for any x ∈ R. If we denote φ(1) = a(> 0), we get that
a solution to our exercise is a function of the form φ(x) = ax , x ∈ R.
4.8. Note the null function is a solution to (4.59). Consider the auxiliary
function
et − e−t
f : R → R \ {−1, 1}, f (t) = t .
e + e−t
It easy to check that function f is one-to-one and onto. Then for any x, y ∈
R \ {−1, 1} there exist u, v ∈ R such that x = f (u) and y = f (v). Moreover,
1 + xy = 1 + f (u)f (v), with 1 + xy ̸= 0.
Consider the function g : R → R defined by g = φ ◦ f. Then (φ ◦ f )(u) +
(φ ◦ f )(v) = (φ ◦ f )(u + v). Since this is precisely the functional equation, we
write (φ ◦ f )(t) = ct, for a constant c. Now we conclude that
c 1+x
φ(x) = cf −1 (x) = ln .
2 1−x
f (x + u + v + xu + xv + uv + xuv)
= f (x) + f (u + v + uv) + f (xu + xv + xuv)
or
f (x + u + v + xu + xv + uv + xuv)
(4.1)
= f (x) + f (u) + f (v) + f (uv) + f (xu + xv + xuv).
f (x + u + v + xu + xv + uv + xuv)
(4.2)
= f (x) + f (u) + f (v) + f (xv) + f (xu + uv + xuv).
or
f (uv) + f (u) + f (v) + f (uv) = f (v) + f (u + 2uv).
42 4 Limits and Continuity
From here
f (0) = 0. (4.5)
Setting v = −1/2 in (4.4), we get f (0) = f (u) + 2f (−u/2) . From (4.5) and
(4.6), we have
From (4.7) and (4.4), we have f (u + 2uv) = f (u) + f (2uv), and by the
substitution 2v = t, it follows
Note that a = f (1) ≥ 0, therefore the set of solutions contains exactly the
functions f (x) = a|x|, for some nonnegative a.
4.11. Define g(x) = f (x) − f (tx), for all x ∈ R. Remark that g is continuous
at 0 and g(0) = 0. We have x2 = f (x)−f (tx)−(f (tx)−f (t2 x)) = g(x)−g(tx).
From
From
x2 t2 x2
f (x) − f (tx) = , f (tx) − f (t 2
x) = , ...,
1 − t2 1 − t2
t2(n−1) x2
f (tn−1 x) − f (tn x) = ,
1 − t2
we have
x2 x2
f (x) − f (tn x) = (1 + t2
+ · · · + t 2(n−1)
) = (1 − t2n ).
1 − t2 (1 − t2 )2
f (x) = k + x2 /(1 − t2 )2 , x ∈ R.
4.12. Exists m > 0 such that |f (x)| ≤ m, for any x ∈ R. Consider the
auxiliary function g(x) = f (x) − x, for x ∈ R. Then g is continuous. From
g(m) = f (m) − m ≤ 0, g(−m) = f (−m) + m ≥ 0, and the continuity of g, it
follows that there exists a ξ ∈ R such that 0 = g(ξ) = f (ξ) − ξ.
4.13. The ”only if” part is clear. Now suppose that (4.61) is true. Then we
have a sequence in a compact interval. It has at least a limit point. Suppose
that the sequence (xn ) does not converge. Then it has at least two limit
points. Choose p, q two limit points with p < q. There must be points from
the interval ]p, q[ in the sequence. There is an x ∈ ]p, q[ such that f (x) ̸= x.
Set ε = |f (x) − x|/2. Then from the continuity of f we get that for some
44 4 Limits and Continuity
δ > 0 for all y ∈ ]x − δ, x + δ [ , one has |f (y) − y| > ε. On the other hand for
n large enough |xn+1 − xn | < 2δ and |f (xn ) − xn | = |xn+1 − xn | < ε. So the
sequence cannot come into the interval ]x − δ, x + δ[, but also cannot jump
over this interval. Then all limit points have to be at most x−δ (contradicting
that q is a limit point), or at least x+δ (contradicting that p is a limit point).
4.14. Consider g(x) = x3 +x, for all x ∈ R. Function g is a third degree poly-
nomial, so it is continuous and onto. Moreover, it is strictly increasing. Then
the its inverse g −1 is strictly increasing and continuous on R, by Theorem
2.20.
Suppose that f is a solution to the exercise. Then
Take y = g(x). Then from the left hand side inequality, we have
From (4.9) and (4.10) we conclude that f (x) = g −1 (x), x ∈ R. Easy to see
that f satisfies the assumption. For, take x ∈ R, y ∈ f (x), f −1 (y) = x, and
g −1 (x) = y. Then g(y) = x and
4.15. Since f is continuous, f ([a, b]) is a compact set. Thus the sequence
(xn ) is included in a compact set and thus it is bounded.
Function f being increasing, the sequence (xn ) is monotone. So we have
established that the sequence (xn ) converges, let x∗ ∈ [a, b] be its limit. From
the estimates 0 ≤ |f (xn ) − x∗ | = |xn+1 − x∗ | → 0 as n → ∞, we conclude
that f (x∗ ) = x∗ .
4.16. (a) Yes. Indeed, let A = {x ∈ [0, 1] | f (x) > x}. If f (0) = 0, then x = 0
supplies the answer. If not, A in nonempty ( 0 ∈ A ) and bounded, so it has a
supremum, say a. Then b = f (a). There are two cases.
I case: a < b. Then, since f is monotonic and a is a supremum, we get
b = f (a) ≤ f ((a + b)/2) ≤ (a + b)/2, which contradicts a < b.
II case: b < a. Then we get b = f (a) ≥ f ((a + b)2) > (a + b)2, a contra-
diction. Therefore we must have a = b.
(b) No. Here is an example
4.1 Solutions 45
{
1 − x/2, x ≤ 1/2;
f (x) =
1/2 − x/2, x > 1/2.
4.18. Suppose that at least one such function exists. Then it is one-to-one
since
Let p1 < p2 < . . . be the set of all primes. It is obvious that f (2k ) = 2k ,
for all k ∈ N∗ , f (3k ) = 3k , for all k ∈ N∗ , and f (2k 3m ) = 2k 3m , k, mN∗ .
mn−1 mn−1
Suppose that f (pm 1 m2 m1 m2
1 p2 . . . pn−1 ) = p1 p2 . . . pn−1 , for all m1 , m2 ,
∗
. . . , mn−1 ∈ N , n ≥ 4. Then pn − 1 are pn + 1 are composed. Their prime
decomposition contain primes from {p1 , p2 , . . . , pn−1 } at nonnegative powers.
Then pn − 1 < pn < pn + 1 implies pn − 1 = f (pn − 1) < f (pn ) < f (pn + 1) =
pn +1. From here one has f (pn ) = pn . Immediately one can see that f (n) = n
for all natural n such that their prime decomposition contain primes from
{p1 , p2 , . . . , pn−1 , pn } at nonnegative powers. Thus f (n) = n, for all natural
n.
4.21. Take an arbitrary but fixed x ∈ R \ Q and two rational sequences (un )
and (vn ) such that un → x, vn → x, and un < x < vn , for all n ∈ N∗ . We
show that f (x) = g(x).
Suppose that function g is increasing. Then g(un ) ≤ g(x) ≤ g(vn ),
f (un ) = g(un ) and f (vn ) = g(vn ). Thus
4.22. From Exercise 4.3 we have that f (p) = p · f (1), for all p ∈ Q. Con-
sider x an arbitrary irrational number, two rational sequences (an ) and
(bn ) converging to x and satisfying an < x and x < bn , for all n. Then
f (1)an = f (an ) ≤ f (x) ≤ f (bn ) = f (1)bn . Passing to the limit, we get
f (1)x = f (x) ≤ f (x) ≤ f (x) = f (1)x. Thus f (x) = f (1)x, x ∈ R. So,
f (x) = ax, for all x ∈ R.
4.23. From |f (x) − f (y)| < |x − y| for all x, y ∈ [a, b], we have that f is
continuous on [a, b]. Inspired by recurrence we define g(x) = (x + f (x))/2,
x ∈ [a, b]. It is easy to see that g([a, b]) ⊂ [a, b]. Moreover, g is increasing.
So the sequence (an ) is bounded and monotonic. Therefore it is convergent.
Let a0 be its limit. Since f is continuous we pass to the limit in an+1 =
(an + f (an ))/2 and find that f (a0 ) = a0 .
4.29. First approach. The set J = f (I) is an interval and f is strictly mono-
tonic, by Corollary 4.9. Suppose f is strictly increasing. Take u, v ∈ J arbi-
trary, but fixed, such that u < v; Then let x = f −1 (u) and y = f −1 (v) be
48 4 Limits and Continuity
their counter images and let λ be a real between them. Since f −1 is strictly
monotonic, we have x < λ < y. Then z = f (λ) ∈ ] u, v [ and thus f −1 (z) = λ.
Second approach. By Corollary 4.9, f is strictly monotonic, while by Theorem
2.20, the conclusion follows.
4.33. It is clear that each term of the series is a discontinuous function. The
partial sum is sn = f /(n + 1). Since |sn (x) − 0| ≤ |f (x)|/(n + 1), for each
x ∈ R, we conclude that the series converges uniformly to the null function.
5
Differential Calculus on R
Now invoking Theorem 1.18 from page 81, we get the conclusion.
(b) Consider the identity
( )
f (βn ) − f (αn ) f (0) − f (αn ) βn f (βn ) − f (0) f (0) − f (αn )
− = − .
βn − αn −αn βn − αn βn −αn
The sequence (βn /(βn − αn ))n is bounded and since the difference in the
parenthesis tends to 0, the conclusion follows.
(c) Invoking the mean value Theorem 2.3, we write
for a sequence (γn ) satisfying αn < γn < βn . Then γn → 0 and the conclusion
follows since f ′ is continuous.
(d) Consider the function given in (b) of Examples 1.1 and the sequences
(αn ) and (βn ) defined as βn = 2/((4n + 1)π), respectively αn = 1/(2nπ).
f (v)
f (w) = lim f (x) = lim g(x) = 0.
x↓w x↓w g(v)
5.3. The inequality is equivalent with ln a/a ≥ ln x/x. So, we are led to
consider the function f (x) = ln x/x, for x > 0. This function has a unique
maximum at x = e. Thus the answer follows for a = e.
5.4. Define g(x) = f (x) · e−x on [0, 1]. Since g(0) = g(1), we find a point
ξ ∈ ]0, 1[ so that g ′ (ξ) = 0. Then 0 = g ′ (ξ) = (f ′ (ξ) − f (ξ))e−ξ , that is
f (ξ) = f ′ (ξ).
5.5. It is clear that all the terms are positive if x1 is so. It is also clear that
if x1 = 0, all the terms are null.
Define the sequence of polynomials
Since
Pn+1 (an ) = Pn (an ) (Pn (an ) + 1/n) = 1 − 1/n < 1 − 1/(n + 1),
Pn+1 (bn ) = Pn (bn ) (Pn (bn ) + 1/n) = 1 + 1/n > 1 + 1/(n + 1),
5.1 Solutions 51
so bn > bn+1 . Thus (an ) is increasing, (bn ) is decreasing, and an < bn , for all
n > 1. Then there exists at least one point x1 satisfying an < x1 < bn , for all
n > 1. Hence 1 − 1/n < Pn (x1 ) < 1, for all n > 1. But Pn (x1 ) = xn+1 . So
xn+1 < 1, for all n > 1. Also xn > 1 − 1/n implies xn+1 = xn (xn + 1/n) >
xn . So, sequence (xn ) satisfies all the required conditions.
It remained to consider uniqueness. Pick an x1 satisfying the given con-
ditions. Then 1 − 1/n < Pn (x1 ) < 1, for all n > 1. So we have for all n > 1
that an < x1 < bn .
The uniqueness of x1 will follow by showing that bn − an → 0, as n → ∞.
Since all the coefficients of Pn (x) are nonnegative, it has increasing derivative.
But Pn (0) = 0 and Pn (bn ) = 1, so for any x ∈ [0, bn ] we have Pn (x) ≤ x/bn .
Then we have 1 − 1/n < an /bn and
n→∞
0 < bn − an ≤ bn − bn (1 − 1/n) = bn /n < 1/n −−−−→ 0.
5.6. The inequality has been already proved geometrically by Exercise 1.51 at
page 39. Here we prove it using Theorem 2.4. We only show that sin x < x,
for every x ∈ ]0, π/2[ . Consider the function f defined by f (x) = x−sin x on
x ∈ [0, π/2[ . Then f (0) = 0 and f ′ (x) = 1−cos x > 0, for every x ∈ ]0, π/2[ .
Hence sin x < x on 0 < x < π/2.
5.7. (a) One might apply directly Theorem 4.1. Instead of it we choose writing
the limit as
( ( 2 ))
xln x ln x
lim = exp lim x − ln ln x .
x→∞ (ln x)x x→∞ x
(c) Since
( )
lim ln(ex + x)/x = lim 1 + ln(1 + xe−x )/x = 1
x→∞ x→∞
we successively have
(√ √ )
ln(ex + x)
x3 + x2 + x + 1 − x2 + x + 1 ·
3
lim
x→∞ x
(√ √ )
x3 + x2 + x + 1 − x2 + x + 1
3
= lim
x→∞
(√ √ )
1 1 1 1 1
1+ + 2 + 3 − 1+ + 2
3
= lim x
x→∞ x x x x x
(1 + t + t2 + t3 )1/3 − (1 + t + t2 )1/2
= lim
t↓0 t
( )
2
1 + 2t + 3t 1 + 2t 1
= lim √ − √ =− .
t↓0 3 3 (1 + t + t2 + t3 )2 2 1 + t + t2 6
5.8. A more general result will be proved, namely f (t) ≥ 0, for all t ∈ ]0, π].
Suppose f (x) ≤ 0 for all x ∈ [0, t]. We show that in this case f (x) = 0
for all x ∈ [0, t]. Indeed, if f (u) < 0 for some u ∈ ]0, t [ , by the mean value
Theorem of Lagrange there exists v ∈ ]0, u[ so that
for some w ∈ ]0, v [ . In this cases (a(w)f ′ (w))′ + f (w) < 0, a contradiction.
Suppose f (p) > 0 for some p ∈ ]0, t [ . Define
ξ1 = sup{x | x ∈ [0, p], f (x) ≤ 0}, ξ2 = inf{x | x ∈ [p, t], f (x) ≤ 0}.
We have 0 ≤ ξ1 < ξ2 < t, f (ξ1 ) = f (ξ2 ) = 0, and f (x) > 0, for all
x ∈ ]ξ1 , ξ2 [ . On the interval ] ξ1 , ξ2 [ define g(x) = a(x)f ′ (x)/f (x). Then
g ′ (x)
g ′ (x) ≥ −1 − g 2 (x) =⇒ (arctan g(x))′ = ≥ −1 (5.2)
1 + g 2 (x)
for x ∈ ]ξ1 , ξ2 [ . One the other side, for arbitrary positive δ there are θ1 , θ2 ∈
] ξ1 , ξ2 [ , with ξ1 < ξ2 , so that
π π
arctan g(ξ1 ) > −δ and arctan g(ξ2 ) > − + δ.
2 2
By the mean value Theorem of Lagrange we find a point ξ ∈ ]θ1 , θ2 [ fulfilling
For sufficiently small δ the last fraction is less than −1, contradicting (5.2).
This contradiction shows that the assumption that f (p) > 0 for some p ∈
]0, t [ is false. Now the conclusion follows.
5.9. From the given equality we may write that for y ̸= 0 we have
f (x + y) − f (x − y)
f ′ (x) = , x ∈ R.
2y
The right hand side is differentiable at x, so
( )
′′ f ′ (x + y) − f ′ (x − y) 1 f (x + 2y) − f (x) f (x − 2y) − f (x)
f (x) = = −
2y 2y 2y −2y
f (x + 2y) + f (x − 2y) − 2f (x)
= .
4y 2
Differentiating again we find
x2
f ′′ (x) = f ′′ (0), f ′ (x) = f ′′ (0)x + f ′ (0), f (x) = f ′′ (0) + f ′ (0)x + f (0).
2
We note that each second degree polynomial f (x) = ax2 + bx + c, x ∈ R,
satisfies all the requirements of the exercise. ⊓
⊔
The limit in the left hand side there exists and we have
5.12. Set
g(x) = [f (x) + f ′ (x) + · · · + f (n) (x)]e−x .
From the assumption one gets g(a) = g(b). Then there exists c ∈ ]a, b [ such
that g ′ (c) = 0. Substituting in the last equality g ′ (x) = (f (n+1) (x)−f (x))e−x ,
we finish the proof.
5.15. Since |e−n sin(nx)| ≤ e−n for every n ∈ N and x ∈ R, the series
converges uniformly and absolutely ∑∞on R. Its sum, denoted by f, is defined
on the whole real axis. The series n=1 ne−n cos(nx) of derivatives converges
uniformly and absolutely on R. Pick up an arbitrary x0 ∈ R and an interval
[a, b] so that x0 ∈ [a, b]. Then by Theorem 7.3, f is differentiable and its
derivative is continuous as a uniform convergent series of continuous functions.
Since x0 is arbitrary on R, the conclusion follows.
6.4. Since all ak > 0, we denote gk (x) = fk (x)/ak . Then for all k,
∫1
g (x) dx = 1. By the arithmetic-geometric inequality (Exercise 1.28 at
0 k
page 35 of the book), we have
∫ √ ∫
1
n
1
g1 (x) + g2 (x) + · · · + gn (x)
g1 (x)g2 (x) · · · gn (x) dx ≤ dx = 1.
0 0 n
√
Then there exists ξ ∈ [0, 1] so that g1 (ξ)g2 (ξ) · · · gn (ξ) ≤ 1, that is
g1 (ξ)g2 (ξ) · · · gn (ξ) ≤ 1. The last inequality is equivalent to
Obviously,
πn
sin y n→∞
−−−−→ 0.
ny 1−1/n 1
For n → ∞, we have
∫ πn ∫ πn
n−1 sin y dy n−1 dy 2(n − 1) n→∞
≤ = (1 − π −n/2 ) −−−−→ 0.
n2 1 y 2−1/n n2 1 y 3/2 n2
Thus ∫ π
lim cos(xn ) dx = 1.
n→∞ 0
First suppose that g takes only positive values. By (h) in Theorem 1.7, we
write for some ξk ∈ [xk , xk+1 ],
∫ xk+1 ∫ xk+1
f (x)g(nx) dx = f (ξk ) g(nx) dx.
xk xk
Further
∫ xk+1 ∫ nxk+1 ∫ (k+1)T ∫ T
1 1 1
g(nx) dx = g(t) dt = g(t) dt = g(t) dt,
xk n nxk n kT n 0
∫ 1
n→∞
f (x)g(nx) dx −−−−→ 0.
xm
Thus
∫ 1 ∑
m−1 ∫ T
1
f (x)g(nx) dx = f (ξk ) g(t) dt + o(1)
0 n 0
k=0
∫ T ∑
m−1
1
= g(t) dt · f (ξk )(xk+1 − xk ) + o(1)
T 0 k=0
∫ T ∫ 1
n→∞ 1
−−−−→ g(t) dt · f (x) dx.
T 0 0
60 6 Integral Calculus on R
Since
∫ xk
(xk − ξk )2 (ξk − xk−1 )2 (xk − xk−1 )2
|x − ξk | dx = + ≤ ,
xk−1 2 2 2
we have
∫ ∑
n ∑
n
b
(xk − xk−1 )2
f (x) dx − f (ξk )(xk − xk−1 ) ≤
a 2
k=1 k=1
1 ∑
n
xk − xk−1 1
≤ = .
b−a 2 2
k=1
Define {
mk , x ∈ [xk−1 , xk [ , k = 1, 2, . . . , n
h(x) =
f (b), x = b.
Then 0 ≤ h(x) ≤ f (x) ≤ 1, for every x ∈ [a, b]. By (6.2), we still have
∫ β ∫ b
ε
0≤ (f (x) − h(x))dx ≤ (f (x) − h(x))dx < .
α a 2
Now, we look for a function g : [a, b] → {0, 1} such that on every [α, β] ⊂ [a, b]
one has ∫ β
ε
(h(x) − g(x))dx < . (6.3)
α 2
Define
1, x ∈ [xk , ηk [ ,
g(x) = 0, x ∈ [ηk , xk+1 [ ,
0, x = b.
Then for each k, ∫ ∫
xk+1 xk+1
g(x) dx = h(x) dx. (6.4)
xk xk
We show that this function g satisfies (6.3). If the boundary points of the
interval [α, β] coincide with some xk ′ , then by (6.4) the left-hand side of
(6.3) is zero.
Suppose there exists an interval [xk , xk+1 ] so that it contains [α, β]. Then
∫ β
b−a
(h(x) − g(x))dx ≤ β − α ≤
α n
∑
p ∫ xk+m+1 ∫ β
+ (h(x) − g(x))dx + (h(x) − g(x))dx .
m=0 xk+m xk+p+1
62 6 Integral Calculus on R
The sum of integrals is zero since each term is so. The sum of the two extreme
integrals is less then ε/2, whenever n is large enough.
We suppose that f (x0 ) ≥ 0 and f ′ (x0 ) ≥ 0, since if f (x0 ) ≥ 0 and f ′ (x0 ) < 0,
we may consider instead of f, the function R ∋ x 7→ f (2x0 − x) and if
f (x0 ) < 0, we consider the function R ∋ x 7→ −f (x). Thus
√
f ′ (x0 ) > 2M0 M2 .
For x ≥ x0 , we have
∫ x ∫ x
f ′ (x) ≥ f ′ (x0 ) + f ′′ (u) du ≥ f ′ (x0 ) + (−M2 ) du
x0 x0
√
= f ′ (x0 ) − M2 (x − x0 ) > 2M0 M2 − M2 (x − x0 )
and
∫ x ∫ x √
′
f (x) = f (x0 ) + f (u)du > f (x0 ) + ( 2M0 M2 − M2 (x − x0 ))du
x0 x0
√ 1
= f (x0 ) + 2M0 M2 (x − x0 ) − M2 (x − x0 )2
2
√ 1
≥ 2M0 M2 (x − x0 ) − M2 (x − x0 )2 .
2
The
√ point of maximum to the right-hand side is attained at x = x0 +
2M0 M2 /M2 . The √ value of the right-hand side on this point is M0 . Then
we get that f (x0 + 2M0 M2 /M2 ) > M0 , which is a contradiction. Thus the
inequality is proved in this case.
Suppose now that p > 2 and 0 < k < p are integers. Denote
1−k/p
ak = Mk /2k(p−k)/2 M0 Mpk/p .
Then
√
ak ≤ ak−1 ak+1 , a0 = ap = 1.
By Exercise 3.18 at page 142, we have that max{a0 , a1 , . . . , ap } = max{a0 , ap }
= 1. Then all ak ≤ 1 and thus the exercise is solved.
6.1 Solutions 63
We note for every real a the function f (x) = x2 + ax, x ∈ R satisfies all
the requirements of the exercise.
6.11. The idea is to write the double sum as an integral and then to use
the Cauchy-Buniakovski-Schwarz inequality for integrals, (i) in Theorem 1.7.
Successively we have
∑
n ∑ n ∫ 1
kl
ak al = kak · lal tk+l−2 dt
k+l−1 0
k,l=1 k,l=1
∫ 1 ∑ n ∫ 1 (∑
n
)2
= kak · lal · t k−1+l−1 dt = kak · t k−1
dt.
0 k,l=1 0 k=1
∑
2n−1 ∑
2n−1
ak ak+p ≥ ak ak+n . (6.6)
k=0 k=0
For every positive ε we can find a rational number p/n such that |p/n − α| <
ε. Suppose n is large enough, at least n > 1/ε. Since f is uniformly contin-
uous, for ε → 0, we have
∫ 2 ∫ 2
f (x)f (x + α) dx = f (x)f (x + p/n) dx + o(1)
0 0
∫ 2 ∫ 2
= fn (x)fn (x + p/n) dx + o(1) ≥ fn (x)fn (x + 1) dx + o(1)
0 0
∫ 2
= f (x)f (x + 1) dx + o(1),
0
pointwise to the function f (x) = 0, x ∈ [0, 1]. The series does not converges
n→∞
uniformly since for xn = 1/n, we have sn (xn ) = 2n/e −−−−→ ∞. On the
∫1
other side 0 f (x)dx = 0 whereas
∞ ∫
∑ 1 ( )
2x n2 e−n2 x2 − (n − 1)2 e−(n−1) x dx
2 2
n=1 0
∞
∑
(1/e−(n−1) − 1/e−n ) = 1.
2 2
=
n=1
∑∞
9 3 xn
= =3 (−1)n n , for |x| < 3,
x+3 1 + x/3 n=0
3
∑∞
−6 −3 xn
= = −3 (−1)n n , for |x| < 2.
x+2 1 + x/2 n=0
2
∑∞ ( )
3x 1 1
= 3 (−1) n+1
− xn , |x| < 2.
x2 + 5x + 6 n=0
2n 3n
∑∞ ( ) 2n+2
1 1 x
2
(arctan x) = 1 + + ··· + , |x| < 1.
n=0
3 2n + 1 n+1
Easily can be checked that the last identity holds also for x = ±1.
6.18. Since the integrand is not defined at x = 1, we ∫have to take into ac-
1
count several subcases. We show that the integral I1 = 0 x ln x dx/(1 − x2 )2
∫1
diverges. But instead of it we discuss the integral I = 1/e x ln x dx/(1 − x)2 ,
showing its divergence. Define f (x) = −2x ln x = x − 1, for x ∈ I = [1/e, √1].
We have f (1/e) = (3−e)/e > 0, √ f (1) = 0, f is nondecreasing on [1/e, 1/ e],
and f is nonincreasing on [1/ e, 1]. Therefore f is nonnegative on I. Thus
−2x ln x ≥ 1 − x, x ∈ I. Immediately we have
∫ 1 ∫ 1
−2x ln x 1 −2x ln x dx dx
≥ =⇒ ≥ = +∞.
(1 − x) 2 1−x 1/e (1 − x)
2
1/e 1 − x
Thus I diverges and hence I1 diverges, too. We conclude that the initial
integral diverges.
6.19. The
∫ a integrand is not defined at x = 1. Consider 0 < a < 1 and
I(a) = 0 (ln(1/(1 − x)))dx. We get I(a) = (1 − ln(1 − a))(1 − a) + 1. The
integral is equal to lima↑1 I(a) = 1, so it converges.
We can approach this exercise by the Mathematica r command.
Integrate[Log[1/(1 - x)], {x, 0, 1}]
(*getting*) 1
6.20. The integral is improper since at 0 and π/2 the integrand might be
undefined. Consider x = sin u. Then
∫ π/2 ∫ 1
sina−1 u cosb−1 u du = xa−1 (1 − x2 )(b−1)/2 (1 − x2 )−1/2 dx
0 0
6.1 Solutions 67
∫ 1
= t(a−1)/2 (1 − t)b/2−1 (1/2)t−1/2 dt = (1/2)B(a/2, b/2).
0
√ √
∥f ∥ ≥ |a1 x01 + · · · + an x0n | = m/ m = m.
Thus ∥f ∥ = m.
For an x of norm one there exists η > 0 such that a + tx ∈ A for all t with
|t| < η. Then
From here we have ∥L(x)∥ ≤ M + ∥ω(a, tx)∥, for all 0 < t < η. Letting t ↓ 0,
by the continuity of ω, we find out ∥L(x)∥ ≤ M. By Theorem 1.3 we write
∥L∥ ≤ M.
7.4. Consider an arbitrary point (x, y) ∈ R2 and define the function g(t) =
f (tx, ty) for t ∈ [0, 1]. Then g is of class C 1 , and
7.6. We show that f ′ (a) = 0, for all a ∈ Rn . Suppose that this is not the
case, i.e., there is an a ∈ Rn so that f ′ (a) ̸= 0. Then for some h ∈ Rn the
scalar f ′ (a)h is nonzero. Admit f ′ (a)h > 0. Pick a real t and for a certain ξ
on the segment defined by a and a + th it holds
xy + yz + zx + 3λ2 = 0. (7.6)
λ2 = 1/6. (7.7)
Subtracting (7.2) from (7.1), we have
xz + yz + λ2 = 0. (7.9)
xz + yz + z 2 = 0. (7.10)
√
√ get that z − λ2 = 0. Thus z1,2 = ±1/ 6.
2
From (7.9) and (7.10) we
(i1 ) Suppose z1 = 1/ 6. Then we have the stationary point
√ √ √ √
M0 = (−2/ 6, 1/ 6, 1/ 6), λ1 = 1/2 6, λ2 = 1/6.
√
(i2 ) Suppose z1 = −1/ 6. Then we have the stationary point
√ √ √ √
M1 = (2/ 6, −1/ 6, −1/ 6), λ1 = −1/2 6, λ2 = 1/6.
We have
∂2F ∂2F 2 ∂2F 2 ∂2F ∂2F
d2 F = dx 2
+ dy + dz + dλ 2
1 + dλ22
∂x2 ∂y 2 ∂z 2 ∂λ21 ∂λ22
∂2F ∂2F ∂2F
+2 dx dy + 2 dy dz + 2 dz dx
∂x∂y ∂y∂z ∂z∂x
7.1 Solutions 73
It is clear that
∂2F ∂2F ∂2F
= = = 0.
∂λ1 ∂λ2 ∂λ21 ∂λ22
From (7.65) by differentiation we get
x dx + y dy + z dz = 0 and dx + dy + dz = 0. (7.12)
We have
∂2F ∂2F ∂2F
2
= 2λ1 , = 2λ1 , = 2λ1 ,
∂x ∂y 2 ∂z 2
∂2F ∂2F ∂2F
= z, = x, = y,
∂x∂y ∂y∂z ∂z∂x
∂2F ∂2F ∂2F
= 2x, = 2y, = 2z,
∂x∂λ1 ∂y∂λ1 ∂z∂λ1
∂2F ∂2F ∂2F
= 1, = 1, = 1.
∂x∂λ2 ∂y∂λ2 ∂z∂λ2
8.3. See Figure 8.1. The two curves intersect at (0, 0) and (1, 1). Therefore
by Theorem 1.6, we have
∫ ∫ 1 (∫ √x ) ∫ 1( √ )
y= x
(x + y)dx dy = (x + y)dy dx = (xy + y 2 /2) y=x2 dx
D 0 x2 0
76 8 Double Integrals, Triple Integrals, and Line Integrals
(1,1)
(0,0)
Fig. 8.1. Figure for Exercise 8.3 Fig. 8.2. Figure for Exercise 8.4
∫ 1 ( √ )
= x x + x/2 − (x3 + x4 /2) dx = 3/10.
0
8.4. See Figure 8.2 and apply Theorem 1.9. We transform the unit closed ball
in a rectangle by x = ρ cos θ and y = ρ sin θ and get
Then
cos θ −ρ sin θ
det T ′ (x, y) = = ρ.
sin θ ρ cos θ
By equality (8.18), we have
∫ ∫ ∫ 2π ∫ 1
π
(x2 + y 2 )dx dy = ρ2 · ρ dρ dθ = dθ · ρ3 dρ = .
D T 0 0 2
8.10. Apply Theorem 2.8. We transform the unit closed ball into a paral-
lelepiped T by
Integrate[Integrate[Integrate[f[x, y, z],
{z, -Sqrt[1 - x^2 - y^2], Sqrt[1 - x^2 - y^2]}],
{y, -Sqrt[1 - x^2], Sqrt[1 - x^2]}], {x, -1, 1}]
∫
8.11. We have to evaluate D
dx dy dz, where D is given as
{(x, y, z) | x2 + y 2 + z 2 ≤ 1}.
Then ∫ ∫ ∫ ∫
2π π 1
dx dy dz = dα · sin βdβ · ρ2 dρ = 4π/3.
D 0 0 0
8.14. The curve is continuous differentiable on its domain and the function
is continuous on the image∫ of the curve. Denote
√ by I the
√ integral. Then after
1 3√
substitution we have I = 0 t 2 + t dt = 8 2/15 − 3/5.
2
γ ∋ (x, y) 7→ 1/(x3 + y 3 )
∫
8.16. We write I = I1 + I2 + I3 , where Ik = γk y dx + z dy + x dz and γk is
the segment Ak Ak+1 , with A4 = A1 . Then γ1 is parameterized as x = 1 − t,
80 8 Double Integrals, Triple Integrals, and Line Integrals
∫
8.17. The area is given by A = 1/2 γ
x dy − y dx, where γ is given as x =
a cos t, y = b cos t, t ∈ [0, 2π]. Then
∫ 2π
1
A= ab(cos2 t + sin2 t) dt = πab.
2 0
There are satisfied the requirements of Theorem 5.3. Then taking the partial
derivatives from I(a, b) in respect to a and b we get
∫ π/2 ∫ π/2
π cos2 x dx π sin2 x dx
= , = .
4a3 b 0 (a2 cos2 x + b2 sin2 x)2 4ab3 0 (a2 cos2 x + b2 sin2 x)2
8.19. We may suppose that a, b > 0. Note that F (1, 1) = 0. There are
satisfied the requirements of Theorem 5.3. Then taking partial derivatives
from F (a, b) in respect to a and b, we get ∂F (a, b)/∂a = π/(a + b) and
∂F (a, b)/∂b = π/(a + b). From
∂F ∂F
dF = da + db,
∂a ∂b
we find that dF = π d(a + b)/(a + b). Then F (a, b) = π ln(a + b) + C, where
C is a constant. Substituting a = b = 1, we get C = −π ln 2. Thus F (a, b) =
π ln((a + b)/2). If we remove the assumption that a, b > 0, then F (a, b) =
π ln((|a| + |b|)/2).
∫b
8.20. Function f is continuous on [0, 1]. We have that f (x) = a xu du, for
all x ∈ [0, 1]. The integrand is continuous on [0, 1] × [a, b], supposing that
a < b. Then we apply Theorem 5.5. Hence
∫ 1 ∫ 1 (∫ b ) ∫ b (∫ 1 ) ∫ b
u du
f (x) dx = x du dx = xu dx du =
0 0 a a 0 a u+1
b+1
= ln .
a+1
We have ∫ b
e−ax (sin(bx) − sin(cx))
e−ax cos ux du = .
c x
The function [0, ∞ [ ×[c, b] ∋ (x, u) 7→ e−ax cos ux is continuous. Therefore
∫ α (∫ b ) ∫ b (∫ α ) ∫ b
−ax −ax
e cos ux du dx = e cos ux dx du = g(u) du,
0 c c 0 c
where
82 8 Double Integrals, Triple Integrals, and Line Integrals